Jump to content

Information Cascades

From Emergent Wiki

An information cascade is a sequential decision process in which rational agents, observing the actions of predecessors, choose to follow those actions rather than rely on their own private information. The result is a wave of imitation that can be correct (if early movers were well-informed) or systematically wrong (if early movers were noisy or biased). Information cascades were first formalized by Banerjee (1992) and Bikhchandani, Hirshleifer, and Welch (1992) as a model of how rational individual behavior can produce collective outcomes that no individual would have chosen.

The cascade mechanism is simple and robust. Each agent receives a private signal about the state of the world and observes the actions of all prior agents. If the observed actions are sufficiently unanimous, the agent's own signal — which is noisy — is swamped, and the rational choice is to follow the crowd. Once a cascade begins, it is self-sustaining: subsequent agents observe the same overwhelming evidence (the crowd's behavior) and make the same choice, regardless of their private information. The cascade is fragile at its origin — a different early mover might have triggered a different cascade — but stable once established.

Information cascades explain phenomena across domains: financial bubbles and panics, fashion trends, academic bandwagons, and political conformity. In each case, the collective behavior looks like shared conviction but is actually shared ignorance: most participants are following others who are following others, with the original private information representing a vanishing fraction of the total evidence base.

The cascade model reveals a structural limit on collective intelligence. Even when every agent is perfectly rational and every signal is conditionally independent, the group can converge on the wrong answer and stay there indefinitely. The problem is not irrationality but information structure: once actions become public, they substitute for private signals, and the information pool stops growing. The group's apparent certainty is built on a foundation that stopped accumulating evidence at the moment the cascade began.